Canonical maps Jean-Pierre Marquis∗ D´epartement de philosophie Universit´ede Montr´eal Montr´eal,Canada [email protected] Abstract Categorical foundations and set-theoretical foundations are sometimes presented as alternative foundational schemes. So far, the literature has mostly focused on the weaknesses of the categorical foundations. We want here to concentrate on what we take to be one of its strengths: the explicit identification of so-called canonical maps and their role in mathematics. Canonical maps play a central role in contemporary mathematics and although some are easily defined by set-theoretical tools, they all appear systematically in a categorical framework. The key element here is the systematic nature of these maps in a categorical framework and I suggest that, from that point of view, one can see an architectonic of mathematics emerging clearly. Moreover, they force us to reconsider the nature of mathematical knowledge itself. Thus, to understand certain fundamental aspects of mathematics, category theory is necessary (at least, in the present state of mathematics). 1 Introduction The foundational status of category theory has been challenged as soon as it has been proposed as such1. The literature on the subject is roughly split in two camps: those who argue against category theory by exhibiting some of its shortcomings and those who argue that it does not fall prey to these shortcom- ings2. Detractors argue that it supposedly falls short of some basic desiderata that any foundational framework ought to satisfy: either logical, epistemologi- cal, ontological or psychological. To put it bluntly, it is sometimes claimed that ∗The author gratefully acknowledge the financial support of the SSHRC of Canada while this work was done. 1The first publication presenting the category of categories as a foundational framework is Lawvere's [23]. As far as I know, the first printed reaction against such an enterprise is Kreisel's appendix in [26]. 2Here are some of the standard references on the topic: [10, 3, 16, 2, 36, 37, 22, 42, 41, 24, 38, 43, 21, 28, 33]. 1 category theory fails to rest on simple notions, or its objects are too compli- cated, or it presupposes more basic concepts, or it can only be understood after some prior notion has been assimilated and, of course, these disjunctions are not exclusive. In this paper, I want to reverse this perspective completely. I want to present what I take to be one of the main virtues of category theory as a foun- dational framework. More precisely, I claim that only a categorical framework, or a framework that encodes the same features than the ones I am going to present here, can capture an essential and fundamental aspect of contemporary mathematics, namely the existence of canonical maps and their pivotal role in the practice and development of mathematics, or should I say their pivotal role in mathematics, period. Moreover, not only can this be seen as a virtue of cate- gory theory, but it should also be thought as constituting a drawback of purely extensional set theories, e.g. ZF(C). I should emphasize immediately that I am not arguing against set theory in general, but only a specific formalization of it. In fact, sets still play a crucial role in a categorical foundational framework3. To be entirely clear: I will not be arguing in favor of a particular categorical framework in this paper. The point I will be trying to make is simply that any foundational framework that is written in the language of category theory necessarily exhibits essential conceptual features of mathematical practice. From a philosophical and a practical point of view, set theory has the ad- vantage of offering an ontological unification: all mathematical objects can be defined or ought to be defined as sets. Thus, number systems, functions, re- lations, geometrical spaces, topological spaces, Banach spaces, groups, rings, fields, categories, etc., can all be defined as sets. This is very well-known. The epistemological aspects of mathematics are supposed to be taken care of by the logical machinery. This also provides a form of unification4. I want to underline that the analysis of mathematics via first-order logic and set theory, which used to be called metamathematics, offers clear and important payoffs. As is also very well-known, category theory itself challenges the ontological and methodological unifications provided by set theory and first-order logic. It is not so much the notion of large categories which is the problem, but rather the inescapable usage of functor categories and functors between them that raises the issue. There are various technical solutions to the problem and we won't discuss them here5. The philosophical limitations of the purely extensional set-theoretical framework are too familiar for us to mention them here6. 3How this can be is explained in [33],[28]. See also [45]. 4Category theory also provides a form of unification, in fact many forms of unification. The most obvious and in some sense superficial is the fact that almost any kind of mathematical structure together with their morphisms form a category. Then, there are deeper forms of unification, forms that are revealed after serious mathematical work has been done. To men- tion but one example, the notion of Grothendieck topos provides a deep unification between the continuous and the discrete, a unification which lead to the development of arithmetic geometry. Our main claim in this paper is based on a different form of unity. The various forms of unification inherent to category theory and their philosophical advantages will be explored elsewhere. 5There is a vast literature on the subject of large categories and set-theoretical foundations for the latter. See, for instance, [11, 12, 4]. 6See, for instance, [29, 35, 27, 34] 2 One of the advantages of category theory is that it puts morphisms on a par with objects, both ontologically and epistemologically. As such, it does not seem to differ that much from one formulation of set theory. After all, Von Neumann has given an axiomatic set theory based on the notion of mapping. (See [46].) What does a categorical framework add to the foundational picture, apart from organizing mathematics differently? Let me start with a loose and informal sketch. Suppose you have laid out in front of view a network of objects with multiple arrows between them. At first sight, the whole thing looks like a messy graph. The fact that stands out from the development of mathematics from the last fifty years or so is that if this graph represents a (potentially large) portion of mathematics, then not all arrows in the graph have the same role nor the same status. When we use a road map or any geographical map, there are conventions underlying the repre- sentation that allow us to see immediately which roads are highways or which portions of the map are major rivers, mountains, etc. The latter representations work because the language used for the construction of these maps contain a code that captures these elements. The point I want to make here is extremely simple: category theory, and not just its language, provides us with the proper code to represent the map of mathematical concepts. To use another metaphor: when the graph is illuminated with the proper lighting, some mappings stand out as having singular properties. It is as if one would lit the given network with a blacklight which make some of the mappings become apparent through fluorescence. The interesting thing is that this can now be reflected in the foun- dational domain and thus acquire a philosophical interpretation. I claim that the illuminated mappings occupy a privileged position: they provide, to use yet another metaphor, the basic routes along which concepts can be moved around. They constitute the highway system of mathematical concepts. This is already significant, but that is not all. Perhaps even more important is the fact that these basic, even elementary, roads open up the way for other conceptually im- portant roads that are in some sense built upon them. Thus, not only do they provide the roads, they also provide the frame upon which the other concepts are constructed or erected. These morphisms are usually called \canonical maps". Once the latter have been identified or recognized for what they are, morphisms that are not canonical but that are important acquire a new meaning too. To use an analogy here, one could say that in the same way that symmetries are fundamental in the sciences, broken symmetries are just as important as long as one has understood the importance of symmetry in the first place. From a set theoretical point of view, canonical maps are merely maps like all the others. They can be defined as sets and they are not highlighted in any special manner. From a categorical point of view, the situation is quite different. The language and the concepts of category theory provide the blacklight. These maps show up as having a special character. They are singled out as being significant. Some mathematicians have decided to use them as much as possible, once they realized the role they played in the architecture of mathematics. Thus, the very practice of mathematics is modified by the conscious recognition of these maps and their status. And their status is not innocuous. Fundamental mathemat- 3 ical theorems rest on the existence and the properties of these morphisms. It then seems reasonable to have a way to reflect this fundamental character in a foundational framework and that is precisely what a categorical framework does. Here is another metaphor: among all the morphisms, some are given right from the start.